Chapter 17
QUESTIONS
1. What is the role of clients and servers in distributed processing?
Ans: In distributed processing, clients make requests to a server. Servers
perform the request and communicate the result to the clients.
2. Briefly define the terms flexibility, interoperability, and scalability. How
does client-server processing support interoperability, flexibility, and
scalability?
Ans: Flexibility refers to the ease of maintaining and adapting a system.
Interoperability refers to the ability of two or more systems to exchange and use
software and data (open standards). Scalability refers to the ability to add and
remove capacity in small units. Client-server processing supports flexibility
because volatile sections of code can be isolated from more stable sections.
Client-server processing supports interoperability because open standards
promote a marketplace of suppliers, leading to lower costs and higher quality.
Client-server processing supports scalability by allowing the movement of
processing between clients and servers and the addition of capacity on the client
and server.
3. Discuss some of the pitfalls of developing client–server systems.
Ans: The pitfalls of developing client–server systems are that it may be more
complex because of architectural choices. In addition to architectural issues, the
designer may face a difficult decision about building a client-server system on
proprietary methods versus open standards.
4. Briefly define the terms scaleup and speedup and the measurement of these
terms.
Ans: Scaleup measures the increased size of a job which can be done while
holding the time constant. Scaleup is measured as the ratio of the amount of
work completed with the larger configuration to the amount of work completed
with the original configuration. With added computing capacity, speedup
measures time reduction while holding the task constant. Speedup is measured
by the ratio of the completion time with the original configuration to the
completion time with the additional capacity.
Chapter 17: End of Chapter Question Solutions
1
5. Briefly high availability and indicate how parallel database processing
supports high availability.
Ans: With highly available or fault resilient computing, a system experiences
little downtime and recovers quickly from failures. Parallel database processing
allows resources to fail without halting processing.
6. How can a distributed database improve data control?
Ans: A distributed database allows the location of data to match an
organization’s structure. Decisions about sharing and maintaining data can be
set locally to provide control closer to the data usage.
7. How can a distributed database reduce communication costs and improve
performance?
Ans: A distributed database can lead to lower communication costs and
improved performance. Data should be located so that 80 percent of the
requests are local. Local requests incur little or no communication costs and
delays compared to remote requests. Increased data availability also can lead to
improved performance.
8. Discuss some of the pitfalls when developing distributed databases.
Ans: Distributed database design issues are very difficult. A poor design can
lead to higher communication costs and poor performance.
9. Discuss why distributed processing is more mature and more widely
implemented than distributed databases.
Ans: Distributed databases are very difficult to implement. A poor design can
lead to higher communication costs and poor performance. In addition, there
are more mature tools in the market for distributed processing than for
distributed databases.
10. Why are division of processing and process management important in
client–server architectures?
Ans: Division of processing is the allocation of tasks to clients and servers. The
load must be balanced for the efficiency of the architectures. Process
management ensures interoperability among clients and servers.
Chapter 17: End of Chapter Question Solutions
2
11. Explain how two-tier architectures address division of processing and
process management.
Ans: A PC client contains the presentation code and SQL statements for data
access; the database server processes the SQL statements and sends query
results. The database server also handles process management.
12. Explain how three-tier architectures address division of processing and
process management.
Ans: Three-tier architectures have the same division of processing as the twotier approach. Middleware is introduced to handle process management.
13. Explain how multiple-tier architectures address division of processing and
process management.
Ans: The additional application servers, which can be invoked from PC clients,
middleware, or database servers, provide a finer division of processing.
Middleware handles process management.
14. What is a thin client? How does a thin client relate to division of processing
in client–server architectures?
Ans: A thin client is a client with inexpensive hardware to handle minimal
tasks for an architecture that has more loads on a server.
15. List some reasons for choosing a two-tier architecture.
Ans: Simplicity. Good for systems with stable requirements and moderate
number of clients.
16. List some reasons for choosing a three-tier architecture.
Ans: Improves performance over two-tier approach. Reduces load on the
database server.
17. List some reasons for choosing a multiple-tier architecture.
Ans: The multiple-tier architecture especially with the generality and power of
software buses and frameworks is the most general client-server architecture. It
can be the most difficult to implement because of its generality especially when
using open standards. However, a good design and implementation of a
multiple-tier architecture can provide the most benefits in terms of scalability,
interoperability, and flexibility.
Chapter 17: End of Chapter Question Solutions
3
18. What is the Web Services Architecture?
Ans: An architecture that supports electronic commerce among organizations.
A set of related Internet standards supports high interoperability among service
requestors, service providers, and service registries. The most important
standard is the Web Service Description Language used by service requestors,
service providers, and service registries.
19. How does the Web Services Architecture support interoperability?
Ans: To support interoperability, the Web Services Architecture uses a
collection of interrelated Internet standards as depicted in Table 17-2. XML
(eXtensible Markup Language) is the underlying foundation for most of the
standards. XML is a meta language that supports the specification of other
languages. In the Web Services Architecture, the WSFL, UDDI, WSDL, and
SOAP standards are XML compliant languages. For a service requestor and
provider, the WSDL is the standard directly used to request and bind a Web
service. A WSDL document provides an interface to a service enabling the
service provider to hide the details of providing the service.
20. Briefly describe the basic architectures for parallel database processing.
Ans: The degree of resource sharing determines architectures for parallel
database processing. The standard classification of architectures are known as
shared everything (SE), shared disks (SD), and shared nothing (SN) as depicted
in Figure 17.8. In the SE approach, memory and disks are shared among a
collection of processors. The SE approach is usually regarded as a symmetric
multiprocessing computer rather than a parallel database architecture. In the SD
architecture, each processor has its private memory, but disks are shared among
all processors. In the SN architecture, each processor has its own memory and
disks. Data must be partitioned among the processors in the SN architecture.
Partitioning is not necessary in the SD and SE architectures because each
processor has access to all data.
21. Briefly describe the clustering extensions to the basic distributed database
architectures.
Ans: In the clustered disk architecture, the processors in each cluster share all
disks, but nothing is shared across clusters. In the clustered nothing
architecture, the processors in each cluster share no resources, but each cluster
can be manipulated to work in parallel to perform a task.
Chapter 17: End of Chapter Question Solutions
4
22. What are the primary design issues for parallel database processing?
Identify the architecture most affected by the design issues.
Ans: The primary design issues that influence the performance of the parallel
database architectures are load balancing, cache coherence, and interprocessor
communication. The clustered nothing architecture is most sensitive to load
balancing because of the need for data partitioning. It can be difficult to
partition a subset of a database to achieve equal division of work because data
skew is common among database columns. By definition, the cache coherence
problem is limited to shared disk architecture.
23. What is the cache coherence problem?
Ans: Cache coherence involves synchronization among local memories and
common disk storage. After a processor addresses a disk page, the image of this
page remains in the cache associated with the given processor. An
inconsistency occurs if another processor has changed the page in its own
buffer. To avoid inconsistencies, when a disk page is accessed, a check of other
local caches should be made to coordinate changes produced in the caches of
these processors.
24. What is load balancing?
Ans: Load balancing involves the amount of work allocated to different
processors in a cluster. Ideally, each processor has the same amount of work to
fully utilize the cluster.
25. What parallel database architecture is supported by Oracle Real Application
Clusters? What is a key technology in Oracle Real Application Clusters?
Ans: Oracle RAC uses the clustered disk architecture. Cache Fusion technology
in RAC enables synchronized access to the cache across all the nodes in a
cluster, without incurring expensive disk I/O operations.
26. What parallel database architecture is supported by IBM DB2 DPF option?
What is a key technology in the DPF?
Ans: DPF uses the clustered nothing architecture. Automatic partitioning is a
key technology of DPF.
27. What is a global request?
Ans: A global request uses data stored at more than one site.
Chapter 17: End of Chapter Question Solutions
5
28. How does the integration level of the distributed DBMS affect the
component architecture?
Ans: A tightly integrated distributed DBMS can access the internal state of
local data managers and thus can be more efficient. A loosely integrated
distributed DBMS, on the other hand, only acts as middleware to coordinate
local data managers.
29. When is a tightly integrated distributed DBMS appropriate? When is a
loosely integrated distributed DBMS appropriate?
Ans: When there is a need to work with legacy systems, a loosely integrated
distributed DBMS is more appropriate, otherwise a tightly integrated
distributed DBMS is more efficient and is a better choice.
30. Discuss the differences in the schema architecture for tightly and loosely
integrated distributed DBMS.
Ans: In tightly integrated distributed DBMS all internal schema formats are the
same. The two new schemas are for fragmenting and locating the fragments.
For loosely integrated distributed DBMS, each site can have different schema
format. A mapping schema is needed to transform data from a local format into
a global format to match the global conceptual schema.
31. How is distributed database transparency related to data independence?
Ans: If database distribution is transparent, users can write queries with no
knowledge of data distribution. In other words, higher transparency means a
higher level of data independence.
32. Is a higher level of distribution transparency always preferred? Briefly
explain why or why not.
Ans: A higher level of transparency is not necessarily preferred. With higher
transparency, users might not perceive the underlying distributed database
processing and could use excessive resource consumption with queries or
transactions. In addition, DBMS support for high levels of transparency may be
inefficient.
33. What is a derived horizontal fragment and why is it useful? What is the
relationship of the semi-join operator and derived horizontal fragmentation?
Ans: A derived horizontal fragment is a fragment defined with a restriction
and semi-join operation. Because some fragments should have rows related to
other fragments, the semi-join operator is important for defining fragments.
Chapter 17: End of Chapter Question Solutions
6
34. What is the larger difference in query formulation: (1) fragmentation
transparency to location transparency or (2) location transparency to local
mapping transparency? Justify your answer.
Ans: Fragmentation transparency is more different from location transparency
(1) than location transparency is from local mapping transparency (2). For some
queries/transactions that operate across fragments, they must be rewritten in a
more complex form when going from fragmentation to location transparency.
However, from location to local mapping transparency, it is just a matter of
adding the site names.
35. Why is fragment design and allocation a complex task?
Ans: Data need to be carefully fragmented and allocated so that most requests
are local. A poor design can cause poor performance and poor resource
utilization. Designing and allocating fragments is similar to index selection.
Data about the frequency of queries, the frequency of parameter values in
queries, and the behavior of the global query optimizer are needed. In addition,
data about the frequency of originating sites for each query is needed. The
originating site for a query is the site in which the query is stored. Just as for
index selection, optimization models and tools can aid decision making about
fragment design and allocation.
36. Why is global query optimization important?
Ans: Global optimization involves data movement and site selection decisions
that could affect response time and resource consumption.
37. What are differences between global and local optimization in distributed
query processing?
Ans: In a centralized environment, minimizing resource usage is consistent
with minimizing response time but in a distributed environment, minimizing
resources may conflict with minimizing response time because of parallel
processing opportunities.
38. Why are there multiple objectives in distributed query processing? Which
objective seems to be more important?
Ans: Because minimizing resources is not the same as minimizing response
time as in centralized processing, both objectives are equally important.
Depending on the application, a designer must find a balance between the two.
Chapter 17: End of Chapter Question Solutions
7
39. What are the components of performance measures for distributed query
processing? What factors influence how these components can be combined
into a performance measure?
Ans: The components are communication costs, local processing costs, and
local input-output costs. Communication costs involved fixed message delays
and variable data transfer delays. The weighting of communication costs versus
local costs (input-output and processing) depends on network characteristics.
For wide area networks, communication costs can dominate local costs. For
local area networks, communication costs are more equally weighted with local
costs.
40. How does two phase locking for distributed databases differ from two
phase locking for centralized databases?
Ans: For distributed query processing, lock management must be done across
all sites. This can be done either by designating a central coordinating site or by
distributing among all sites.
41. Why is the primary copy protocol widely used?
Ans: The primary copy protocol reduces overhead at the cost of a non-current
secondary copy. Usually performance is more important.
42. What kinds of additional failures occur in a distributed database
environment? How can these failures be detected?
Ans: The new failures are communication related. They can be detected by
coordinating among sites.
43. What is the difference between the voting and the decision phases of the
two phase commit protocol?
Ans: In the voting phase, the coordinator writes Begin-Commit and waits for
the participant to send in a response (ready or abort). The decision phase begins
when the coordinator either receives votes from each participant or a timeout
occurs. If a timeout occurs or at least one participant votes to abort, the
coordinator sends abort messages to each participant. Otherwise, the
coordinator sends commit messages to each participant and waits for responses.
44. Discuss the trade-offs between centralized and distributed coordination in
distributed concurrency control and recovery.
Ans: Centralized coordination is simpler than distributed coordination but can
be less reliable.
Chapter 17: End of Chapter Question Solutions
8
45. What level of transparency is provided by Oracle distributed databases?
Ans: Oracle supports local mapping transparency. With the usage of synonyms,
Oracle supports location transparency. With the usage of views, Oracle
supports fragmentation transparency.
46. What are database links in Oracle distributed database processing?
Ans: A database link provides a one-way connection from a local database to a
remote database. A local database is the database in which a user connects. A
remote database is another database in which a user wants to access in a global
request. Database links allow a user to access another user's objects in a remote
database without having an account on the remote site.
Chapter 17: End of Chapter Question Solutions
9